The Telegram channels of Iraq’s armed factions and their media networks, collected continuously and read against a strict attribution taxonomy. Threats, calls to escalate and attack claims are separated from rhetoric; silence from an actor that normally speaks is itself reported.
The signals →About the desk
Iraq Monitor watches Iraq from the sources themselves: what the actors say in their own words, what the state records in its own documents, and what the numbers do. It publishes what it can evidence and says plainly where the evidence runs out.
What the desk watches
Thirty-odd series carried from the institutions that publish them — the Central Bank’s monthly bulletin, SOMO’s export tables, the Iraq Stock Exchange, WFP market monitoring, Brent. Parallel and official exchange rates are shown side by side, because the gap between them is the story.
The data room →The Federal Board of Supreme Audit publishes 713 pages of Arabic fiscal tables each year: the state’s own record of where its money went. The Audit extracts them, verifies every figure against the report’s printed totals, and names the findings to institutions.
The Audit →Every claim is placed on the map with its status — claimed, denied, reported, stated — and never a verdict. The record then counts, source by source, how much of what each one says independent reporting ever bears out.
The claim record →The rules
Primary sources are the basis of a finding. The actors’ own words, the state’s own documents. External reporting and research appear labelled and are never the basis of a finding on their own.
A claim is a claim until it is corroborated. Attribution is quoted or it is not asserted. A statement relayed by a channel belongs to the party quoted, not the channel.
The attribution taxonomy is carried into everything. The desk distinguishes an official armed-group channel from a leader’s channel, an affiliated outlet from an amplifier, and a façade brand from the group assessed to operate it. Every published item inherits those distinctions.
Nothing publishes without clearing the gate. A second model reviews every claim against the evidence behind it. A claim reaches the site only if that review finds it supported, it meets the desk’s evidence threshold, and it links to the source material it rests on. Claims that fall short are held for a human expert rather than published with a hedge. Experts also read published items at random, and correct or withdraw them in the open.
The desk is taught, and the teaching is kept. Human experts with deep subject knowledge train the system on verified answers — the correct reading of a claim, the right actor, the true status of a source. Every correction is stored and fed back, so the desk operates under a growing body of rules drawn from real editorial judgement rather than the model’s own assumptions. It becomes more accurate over time, not merely faster.
Corrections are made in the open. When something is wrong the item is amended or withdrawn visibly, carrying the correction with it. Nothing is quietly deleted.
How the claim record is measured
What counts as a claim. Only assertions that could have left an independent trace: a strike on a named place, a casualty figure, a seizure, a named person’s arrest or death, a deployment. Statements of intent, threats, grievances and predictions are recorded but never scored — nothing outside the channel could confirm or refute them, so counting them would only manufacture a number.
The test is written first. When a claim is entered, so is the specific thing that would falsify it — before any search is run. That is what stops a check drifting into finding whatever the searcher expected.
The search is over non-militia sources only. Iraqi and international press, official statements, monitoring bodies. A claim repeated by three allied channels is one claim, not three confirmations, and never corroborates itself.
Checks repeat on a ladder. Independent reporting lags, sometimes by days. A claim is re-checked at widening intervals rather than judged once, and counts as checked only when the ladder has run out.
Claims about places this desk cannot see are never rated. The independent record is built from reporting inside Iraq. When a group claims a strike on Eilat or eastern Syria, Iraqi outlets report the announcement, not the event — so scoring it would measure this desk’s reach rather than the claim. Those claims are collected, published and counted, and kept out of every rate.
No rate below 10 checked claims. Under that floor the counts are shown and the rate withheld. “1 of 2 confirmed” is not “50% reliable”, and a percentage travels further than the sentence qualifying it.
Untraced does not mean false. It means no outlet outside the militia information space reported the event either way. Iraqi coverage is uneven, whole districts and weeks go unreported, and these channels are sometimes genuinely first. Only the contradicted column carries a finding against a source.
How the Audit is verified
Every figure is transcribed twice by machine, in two independent passes reading the page in opposite directions, then compared cell by cell. Any disagreement stops publication. The rows are then summed and reconciled against the total the page itself prints; a table that does not add up is withheld, not shown with a caveat. Numerals are converted by code, never by judgement, and every published figure carries the printed page it came from.
Audit observations are the Board’s recorded findings, not convictions, and they name institutions, never individuals. English renderings of Arabic names are the desk’s own and are labelled as such; the Board’s Arabic sits beside them and is the authoritative text.
Iraq Monitor is independent. Every claim links to its source; nothing here is a verdict.